Subject:     G3/iMac 4MB VRAM WARNING!!!
Date:        10/23/98 3:48 AM
Received:    10/23/98 7:55 AM
From:        Kent Kanja,
To:          Mike,

Mike,

I recently purchased a 4MB VRAM from LLB.  It just so happens that the VRAM
they shipped me did not work in my Rev.2 PowerMac G3 266 (Rage Pro).  I
don't know, nor am I claiming to know how it would operate in other
PowerMac G3's or iMacs, but since they use the same VRAM modules (and now
the same video hardware) it is likely that they would not function either.
This only affects the highest VRAM consuming resolution/video depth
combination.  1280x1024, Millions of colors.  And only affects the lower
1/5 of the screen - however it is still a problem and, personally I like
that resolution on my 19" monitor.  It also only manifests with the ATI
Graphics Accelerator (2d graphics acceleration extension) installed.  I
tried 2 different versions of the extension, the first (rev. unknown, came
preinstalled).  The second is the latest available from ATI.
The problematic VRAM module is manufactured by Micron - LLB's item number
for it is 1520SG "4MB VIDEO SGRAM FOR LEGACY."  The module itself has
stickers with the number MT4LG51264HG83 on it.  The surface mount chips
themselves have the number MT41LC256K32D4 on them.
LLB has shipped me 2 replacement modules, the first replacement was a
similar VRAM (same numbers, same manufacturer) and also failed to work
properly.  The second replacement has "SGS" silkscreened on the edge of the
module, and works just fine at all resolutions and color depths that I can
support.  (Up to 1920x1080 thousands of colors).  I made sure to check the
1280x1024 millions first, of course...  (In my eyes, Bill, my contact at
LLB, has redeemed his company's name - he made sure to send me the VRAM
made by a different manufacturer)

Wow, that was long winded.

Okay, I've taken up enough time - if you want this electrical engineer's
analysis of the problem, it's below.
Kent Kanja (MiM)


Here's my analysis of the problem:
It is likely that the address lines to the highest 2MB of VRAM were not
reaching a stable state when the ATI hardware was trying to write to the
VRAM.  The problem only occured with (some) accelerated 2D VRAM access,
such as drawing rects and inverting text.  This explains why disabling the
ATI Graphics Accelerator extension would prevent the problem from occuring.
Dragging an affected window into an unaffected portion of the screen would
copy the corruption along with it, indicating that it was not a VRAM read
access problem.